A nonparametric method for the segmentation of the EEG.

A new method for segmentation of the EEG, based on a nonparametric statistical analysis, is proposed. A nonparametric approach was chosen because it minimises the need for a priori information about a signal. The method provides detection of change-points (quasi-stationary segments' boundaries) in almost any EEG characteristic for a given level of false alarm probability. The method was applied to 8-channels spontaneous EEG recordings obtained from 12 subjects in eyes closed and eyes open conditions to detect rapid fluctuations of the alpha rhythm power. After preliminary adjustment of false alarm probability values all the recordings were analysed in unsupervised regime with the same parameters. From 15 to 119 change-points were found per minute and EEG channel. Automatically detected change-points were in good correspondence with visual estimation of the instants of change in alpha activity.